ApiaryActiveLive
Try: pause · settings · learn · wipe
← Community / Reading Room
AT
craft · 24 min read

Approve the First Piece, Not the Whole Batch

Twenty-four signs are due Friday. The design is simple: a small paper placard with a dark border and a space for a holder. On Thursday, somebody looks down at…

Twenty-four signs are due Friday. The design is simple: a small paper placard with a dark border and a space for a holder. On Thursday, somebody looks down at the stack of printed placards and notices the border shifted left. Was it on the proof? The proof had no border. The editor added it after the approval email. Did the customer approve the trimmed version? They said “Looks good!” under a screenshot of three layout options. Is the print run released? Nobody knows exactly which thing was released. They only know someone said yes.

A first-piece release is an answer to a better question: does one identified result from this identified setup give us enough evidence to run the work described in this revision? The receipt should say what was observed, who decided, and what the decision permits. It should explicitly leave untouched things the sample cannot prove. A first piece can check that the process is pointed in the right direction. It can't show that number 17 had the same dimensions after a fence shifted or a stack drifted.

The example in this guide is a fictional small shop making non-structural paper display pieces. It isn't a plan for medical items, safety equipment, engineered components, or regulated production. Follow any actual specifications, required inspection plans, and approvals that apply to your work. Borrow the separation of decisions, not a sector's formal inspection rules or legal language.

Four different things somebody might call a sample

A concept sample answers whether people want a thing like this. It can be crude, the wrong material, and wildly unlike the output other than the one feature you want to show. Put a big label on it: “look and layout only” or “prototype; not production.” It's useful at precisely the point where making it production-like would waste time.

A proof answers whether content, placement, or selected appearance is correct in an identified representation. A proof ordinarily doesn't replicate all production details. A PDF can settle spelling without proving color under a particular light, or whether the trimmed border lands where you want it. A physical proof and a digital proof might both be correct for their stated job. When somebody approves it, keep the asset or a reliable version pointer. “Yes to the proof” is nearly useless when you can't find it again.

A first piece comes off the setup you intend to use, with the closest practical match to the run conditions prescribed by the job. Inspect the functions that particular setup can affect. A sign cut from a duplicate digital layout may look perfect before the fold. The first piece, after the ordinary folds and trims, can tell you something the layout cannot. If the first piece is made on alternate paper or a different printer, state the difference and don't claim it demonstrates the former setup. As a deliberate choice, start with a first piece made by the production process or make a narrower decision describing what remains pending.

A retention sample represents something you agree to keep for later comparison. It might be the very piece you examined for release, but the retention label is a different fact. The same object, carefully stored, gives tomorrow's team a shared reference; it won't magically stay pristine or unchanged in storage. An old sample beside a revision change is a strong temptation to transfer yesterday's decision to today's job. Label the relationship so explicitly that visual correspondence alone can't be mistaken for authorization.

ASQ's quality glossary discusses a sample in the specific context of drawing from a lot, and describes using a standard such as a physical product sample for comparison. Those two uses matter here: an early setup piece is intentionally not a random member of an existing completed lot. The same object can be excellent for the first purpose, and unable to do the second.

Write down what's important before you see anything

A useful check can fit in a short list. It starts with details that can actually vary. “Does the sign look good?” gives a single observer too much latitude to pass what a second observer would reasonably call wrong. “Finished width 125 millimeters, allowed example band 124 through 126; holder tab must slide into fixture H-02 without folding; all lines of wording match proof R4” offers three independent decisions. If all are true, write that. If one doesn't apply to this revision, say “not applicable” and why. Don't turn the row off simply because it was expensive to check.

Avoid taking every detail in the design file as equally important. A decorative corner mark might live happily with a sliver of variation. The tab that holds the placard in position may need more scrutiny. Read the specification, ask what the part touches, and differentiate fit, appearance, legibility, and content. Choose observations that match each. A ruler cannot tell you whether the words are right. A photo that looks crisp can't tell you whether the physical sample sits flush. A scanner or a better camera isn't always the answer. Sometimes two people need to agree on the physical comparison and view it under the same conditions.

For compare-by-appearance work, agree what reference to use and when. A warm lamp changes the viewing conditions without changing the sign. So might a glossy sleeve or a dim shelf. The first-piece card could say “compare unlaminated front color to approved sample P4 beside the window under ordinary diffuse daylight; don't read a color match from phone pictures.” That is a local instruction, not rigorous color measurement. If a quantified color requirement matters, get the proper standard and measurement approach, and keep this card as the pointer to that check rather than replacing it.

Specify where tolerances and requirements came from. A friendly guess about what probably fits is not a composition rule for the run. If the drawing says one thing and the approved proof another, mark the mismatch and hold that feature before collecting a row of pretend passes. Treat a blank as “not decided” or “not checked,” not as permission to accept everything. Often the little card makes a problem visible earlier than the inspection itself. Finding that the team has never agreed on the sign's finished width is the sort of result that saves paper.

Make the release narrow enough to trust

The sample card identifies which unit was observed. “One of the blue signs” won't do when they are all blue. Mark the label or sample container prominently, without altering the usable face. Use the job ID, revision and a sample sequence. If you make multiple attempts, keep them distinct: A failed tab fit; B was measured after the stop moved; C was abandoned when the drawing changed. There is no practical reason to overwrite A with C. Start a new record when you change the condition that the earlier claim described.

Tie the record to the production setup. Record the equipment or station if interchangeable shops matter, the template filename and revision, and consumable or material details with mechanisms that can change the outcome. Don't list the operator's entire inventory on every job. For paper signs, stock thickness, layout file, cutter stop, and holder fixture might matter. A shortcut such as “same as previous” is only useful when other people can find the exact previous conditions and know which variables it implies. Include effective dates or a run identifier when different jobs can coexist on the floor.

Name roles rather than treating “approved” like gravity. The customer might approve the content proof; the floor lead might approve the setup from the physical sample; a closer might perform the agreed final checks. One person may fill all three roles on a one-person job, but write down which question they were answering. The buyer's approval of the spelling does not automatically empower whoever saw the message to release the run. Likewise an operator's check on the fixture can't decide which customer logo is correct. You can default to one named decision owner for each release and have that owner ask others for information rather than circulate ambiguous “thoughts?” requests.

The release itself is a controlled handoff. State the revision, permitted quantity or next step, material and condition covered, remaining checks, exceptions, and hold triggers. A quick message can provide this if it is readable and stored next to the job. A required signature system may apply in some contexts; use it if it does. The example does not need one. What matters is that the decision can be recovered and interpreted by somebody who wasn't in the room.

Worked example: three decisions on the same sign

Everything here is fictional. One little archive wants twenty-four table labels for a weekend event. The content was approved on PDF P4 at 10:20. The production master M4 has the same wording as P4, plus the verified trim location prescribed for the job. Only one holder fixture fits this run. The lead makes sample A on the normal paper, after the ordinary trim, and records the result:

CheckSample ARule for this exampleResult
Finished face width125 mm124–126 mmPass
Holder slot fitTab folds as it's insertedSlides and sits without foldingHold
WordingMatches approved P4Exact wording and line orderPass
Border positionLeft 4 mm, right 4 mm3–5 mm at the two named measurement pointsPass

The release cannot be “three out of four.” If the installer cannot insert the sign without bending it, good margins do not compensate. Rechecking the wording does not address the hold. The operator changes the setup after finding the tab's projection was wider than intended. They record what was changed, make sample B, and verify its width, fit, wording and border. A new sample gets a new record; keep the first one visible in the same job because it explains why the setup is no longer the one used at the start.

Possible final entry: “Job W24, master M4, sample B, stock code S2, cutter setup C2, holder H-02. First-piece checks complete: 125 mm overall, clean fit, wording matches P4, margins 4 and 4 mm. Release 24 labels from C2 using S2. Hold on stock, master, holder, or cutter change. At close, inspect all wording and seating as agreed. A failed; never mix it into the run.”

Notice the small amount of new content in the release. No reopening the quote, no approval of extra materials, no filing away a big PDF full of facts nobody checks. It draws a line from the reviewed evidence to the next permissible action. If a failure shows up later, W24 and C2 tell you which bundle of evidence to reopen.

A photo can supplement the record if somebody finds the physical sample later or a colleague needs a check before entering the shop. Put the sample ID beside the piece in one view and show the point measured in another only if necessary. Don't present pictures as independent proof of a fine dimensional reading or as substitute evidence for a fit test. The log can describe exactly what the inspector felt or saw without claiming a screenshot lets another person repeat the conclusion. If the fixture has an ID, capture that too. The image should carry the facts necessary to identify the object, not just make it look convincing.

Where premature confidence likes to hide

The sales person sends a comment as “approval,” but it refers to a social image, not the print file. The operator chooses the prettiest piece from several attempts and calls it the first, forgetting to mention that the other attempts behaved differently. The person evaluating fit flexes the piece just enough to make it work without recognizing that the agreed test prohibited bending. The release form lists the intended material, but the only stock on the bench came from a different box. These are mostly good-faith acts. They all break the chain between the decision and the conditions described.

Guard against that with a short set of triggers. Resample after new work conditions that reasonably affect the result. If the cutter is adjusted, the reference fixture exchanged, or the layout file replaced, the old release remains a true record of old conditions but no longer covers the modified ones. If a truly identical replacement is planned under a validated shop procedure, follow that procedure instead of declaring any unchanged line impossible to use. In a small shop without that evidence, making one more piece is often cheaper than debating whether two nearly identical files are interchangeable.

Keep final inspection visibly separate. Inspect all units when that is what the work calls for; where a documented lot-sampling procedure applies, follow that. The NIST discussion of lot acceptance sampling describes assessing evidence from a randomly chosen sample to make a decision about a lot. That is not what sample B did. Treating a good first piece as a random sample from future output is a category mistake. Conversely, finding a bad twentieth piece doesn't retroactively make B dishonest. It prompts a look at what changed and what work lies between the last useful inspection and the new discovery.

Take special care with residual hold items near the bench. If the first or failed pieces do not count toward ordered quantity, mark and store them so someone won't accidentally pack them. Keep records long enough to resolve the problems they were meant to clarify and in keeping with requirements that apply to you. If you show a retained piece to train the next shift, say whether it is a good example or an instructive bad one. A box full of anonymous samples buys nobody confidence.

The shortest version that still works

If you can get only one part of this into everyday practice, ask three things before using the word approved: exactly which object was checked; against which version and conditions; approved for what next action? Answers can be three fields in a job ticket or one line over a tray. Keep spelling, fit and finished quality separate whenever the same “yes” could be misread to cover all three. In a lunch break someone can write “proof accepted; first piece pending.” That sentence is cheaper to read than a return trip to the printer.

Formal production contexts sometimes set out first-article approval rules very explicitly. For example, the U.S. Federal Acquisition Regulation's first-article subpart defines approval in terms of acceptance of test results, distinguishes first-article review from the production quantity, and discusses avoiding premature production. It is an example of the distinction, not the shop's contract and not a legal template. A paper sign should not need that machinery. It should still have a record that tells you what the yes meant.

Write a decision someone can reproduce without the approver present

Imagine a school exhibit committee that wants a dozen small cardstock labels. The job is deliberately low risk: the labels sit in existing tabletop holders and do not serve as emergency directions, legal notices, or structural supports. The organizer approves the wording and placement on an early draft. The maker prints a first item using the intended stock and holder. The first item could invite two very different questions: does this type look right from the display table, and did this exact setup produce the placement we agreed to use for later copies? Put the two decisions on different lines. Someone who likes the style may not have checked the process, and someone who verified the position need not be approving a late wording change.

List the approval conditions before the first piece comes off the setup. In the fictional case, use the final label text, the selected stock, the currently chosen holder, and a known folder revision. To assess readable placement, review the printed line from the planned display position. To assess fit, use the actual holder intended for that exhibit rather than an unrelated mock piece. If the committee expects the orientation arrow to face the viewer, identify exactly which side is the reference. None of this requires buying professional inspection apparatus. It does require describing the question well enough that two competent reviewers looking at the same item can agree what they just checked.

Then give the first piece its own identity. A local tag or a reference photograph might suffice. Write whether it came from the current setup or was a sample borrowed from the prior arrangement. In a small classroom project, somebody could bring an older item to show the intended look. It is quite useful as a comparison standard, but it doesn't prove the new setup made it. That distinction is close to the difference between a standard used for comparison and a sample selected from a lot discussed in the ASQ glossary. A clear label lets the workshop use both pieces without assigning either more evidence than it has.

Try a synthetic handoff. The original approver isn't there on the next morning. Another staff member has the job revision, the intended holder, the first-piece label and its approved setup record. Ask that second person to explain what can proceed and what would force a new preview. If they cannot tell, the approval is not finished no matter how clear it felt in the room yesterday. The best first-piece card reads as a scope statement, not an invitation to guess whether everything that follows carries the earlier approval. This is especially important when the maker expects the next phase to happen during a different shift.

A minimum decision line can say “First piece E3 reviewed against wording revision R2, stock P and Holder H; permission to make at most twelve matching labels in this setup; count and appearance verification still due after production.” Use the real roles and local limit agreed for the pilot rather than the invented number here. If the job grows, the team might update either the allowed run or the entire release scope; don't mark fourteen pieces accepted just because twelve had been authorized. The approval was a permission to begin controlled work, not a claim about output no one had seen.

Keep sample approval tied to the conditions that made it meaningful

A setup includes more than the name of a job. The relevant details are the parts of the process that can change the expected output or the way the reviewer interprets it. In the labels example, scaling and orientation on the printer, selected source revision, stock substitution, and type of holder can all matter. Room temperature doesn't need a dedicated field just because some other production work logs it; the procedure should include what its evidence uses. Conversely if the physical appearance is the decision, lighting and viewing angle can matter even where the printed dimensions don't change. The release card can describe an ordinary review condition without implying laboratory precision.

Suppose the first piece holds its text clear of the lower rail. The maker later replaces the stock with a thicker material, and the label sits differently. A careless reviewer says same wording, same approval. The synthetic test can show why that is insufficient. The permission covered a combination of stock, holder and setup. When a specified input changes, pause only the affected portion and ask whether a new first piece is needed. Do not broaden the scope of approval from an unchanged title. A small release card gives the team a quick route to a fair decision instead of a broad command to redo all its work.

Try the opposite case. A corrected worksheet changes an internal job nickname but does not alter the selected revision or any specified output condition. The change might require updating traceability but not a fresh sample. A useful policy identifies which fields matter to reproduction and which are administrative. This distinction prevents a process that either lets big changes slide or slows ordinary work on harmless edits. Link the no-change explanation to what controlled the artifact, not simply to the person's confidence that the new revision will look fine. A no-change decision is itself reviewable evidence.

When changes happen in the middle of a batch, find the boundary that owners can defend. Imagine that five labels were made under the reviewed setup and two after a known switch. Mark the change interval as exactly as your evidence allows. Keep the already reviewed output receipted separately from suspect output, and avoid re-labeling everything as the same approved run until the affected pieces can be evaluated. If no record establishes when the switch happened, the uncertainty must be carried into the review rather than reconstructing a precise break from dusty scraps or filenames. Make ordinary replacement and mixed runs explicit choices, not accidents in how objects are counted.

A compact rehearsal checks three cases: a change clearly out of scope, a change clearly irrelevant to the output, and a change with unknown impact. The first should lead to a held release for that setup. The second can continue with a dated annotation if permitted by the house method. The third needs review before new work relying on the original permission. The test doesn't ask someone to produce unapproved objects in a live job. Use paper records and known sample pieces from your own fixture. Good tests catch ambiguous permission without turning a routine workshop into a bureaucratic project.

Handle a review whose answer is neither an unconditional yes nor an erasure

In the fictional exhibit, an organizer approves layout and stock but asks for a corrected date before the first production copy. That is not an authorization to print copies while the edit is pending. The release card can say layout accepted at R1, production blocked pending confirmation of exact final date at the next revision. When the change arrives, review the parts the conditions call out and the consequences of the changed content. A longer date line can change wrapping even if the style remains approved, so a selective recheck helps. Keep geometry and content evidence connected without forcing the team to pretend that one evidence item settled the other.

If the corrected label prints successfully, identify the new first piece and state which earlier review it's using. It may reuse the approved stock and holder while superseding the old line-break decision. A short note can establish that relationship more honestly than crossing out the existing record until it looks like the R2 label was reviewed days ago. Preserve the earlier decision and time because it explains why a staged piece exists or why the maker did not run the batch. If someone reviewing the job is confused about which later object was approved, the card should have a direct answer without reading raw chat history.

A conditional run can be useful when production naturally proceeds in phases. For the simple labels, the organizer can approve printing the first group for the near shelf but not release the distant shelf labels whose content differs. Those are known separately labeled groups, not measurements of the same sample. The maker may also need a second item for a fit check before a repeated run. Whatever permission exists, use an explicit state, allowed quantity or scope, limitations, and named next evidence. A signoff that says “OK with changes” transfers the unsolved change identification to the least informed person in the room.

It is possible to reject a first piece without condemning the whole job. A rejected sample might reveal a reversed arrow using the otherwise intended process. Record which criterion it failed and which reviewed conditions remained unaffected. It is rarely useful to mark all prior thinking invalid without explanation. Change the smallest thing that makes an appropriate test possible, then identify the next piece and repeat the relevant checks. This also prevents a normal first-piece iteration from looking like a mysterious quality incident in a report. Nobody needs to hide a useful failed trial to show the final batch was prepared correctly.

Include the quiet unresolved outcome. The person permitted to approve sees a problem but cannot establish whether it was in a source revision or the setup. They can hold the start and ask for a comparison using the retained job facts. Do not have an inexperienced replacement infer a yes from the absence of a written rejection. The short public status can say waiting for first-piece decision, with a work-owned next step, without sharing a person's private contacts or histories. This gives a workshop an honest flow even if the reason for the hesitation isn't known at the moment of handoff.

Make the final comparison answer a different question

Once the fictional labels are produced, a review can confirm the intended group is complete and check its actual visible properties. This is a different decision from whether an initial piece matched the setup. For a small, simple set the agreed final procedure may inspect each item. If an organization legitimately uses sampling against a documented plan, follow that plan, not an improvisation based on a handful that happened to be nearest the drawer. The NIST discussion of acceptance sampling locates that concept in the decision about a lot from samples under a design. An early setup piece hasn't been selected from future completed output.

In one synthetic failure, the committee finds that the ninth and tenth items have a shifted footer. The first reviewed piece is still genuinely correct. The failing later items may reflect a setup change, a lost orientation rule, or a nonrepeatable condition. Record the observed discrepancy and authorized next action. A process review might identify an affected range from reliable evidence, or need to check more pieces because the range is unknown. Keep condition notes about output separate from blame. The receipt should make reliable repair possible rather than turning the initial approver's name into the automatic explanation for later evidence.

Build a distinct comparison case in which every later label has the exact footer location, but there are only eleven in a twelve-label order. First-piece approval can neither account for missing items nor verify where each object will be placed. Count and final disposition are still open. The workplace may intentionally make an extra test label or retain the approved sample, so define which pieces belong to the promised delivery inventory. A report of twelve physical objects doesn't automatically mean twelve finished usable labels are present. Use object identity and status where helpful instead of assuming all pieces were equivalent when counted.

Another case concerns change at the moment of delivery. A site volunteer asks to swap two already made labels because a different drawer arrangement looks nicer. This change might be completely acceptable, but it needs the right decision about final placement. It is not proof that the first-piece setup was wrong or that the run was already accepted under the new display rule. A short post-run amendment can say which objects changed placement without reusing the original approval as a blank endorsement. The more clearly each decision is scoped, the easier it is to approve reasonable changes without falsifying the provenance.

For an operational handoff, preserve one held item, one correctly produced item and the revised first piece in a paper-only pilot. Ask a second reviewer whether the workshop has permission to produce the next group, whether the existing lot is ready for use, and which evidence would change their answer. The same artifact might support more than one question, but the answer should explain when. This exercise applies in small low-risk production without pretending to replace formal quality programs, legal criteria, or appropriately qualified inspection for critical work.

Retire an approval when its assumptions are gone

A team might later find the retained sample in a cabinet and decide to run a second batch. The artifact can show an earlier appearance, but the original approval may have ended after the allowed quantity, a particular shift, or a modeled setup change. A new order can reference it as a comparison item without claiming continuing permission. Record the new job revision and ask which inputs can be verified again. If the old card leaves out enough information that the process cannot be reconstructed, treat it as historical, not as a portable release method.

Try a small envelope labeled keep only for comparison. It holds the approved label and a copy of the first-piece card. The next person can see how to identify the job, the checked reasons, what the earlier allowed run included, and whether the paper itself needs special retention. A fresh comparison can be performed against this object if appropriate; its presence never bypasses current release authority. It also keeps two obligations separate: preserving evidence of prior work and establishing whether the new work meets today's requirement. One artifact may be sufficient for the first and not even relevant for the second.

If a more formal process applies, do not port these informal examples into it as legal or contractual advice. The Federal Acquisition Regulation's first-article subpart is evidence of how consequential production boundaries can be set in one formal context; the examples here remain private shop decisions for benign display work. The transferable principle is to describe what the evidence warrants and who is authorized to make which decision. A simple card pays for itself when an ordinary shift leader can tell whether they're looking at an approved setup, a useful model, an unfinished group, or a delivered order.

Try a reviewer handoff under two neighboring job names

Small shops can personalize two orders that look almost identical. Imagine that the fictional display has a Monday version and a Tuesday version, each with the same exhibit heading and nearly indistinguishable stock. A maker opens the right template but clips the wrong first-piece card to the holder. Check which fields distinguish the artifacts without trusting the conspicuous heading. A reversible paper exercise can mix two synthetic cards and ask a reviewer to compare the local job ID, exact wording revision, physical label, and intended run quantity. The wrong card should fail the join even if the paper has the correct fit and general style.

Now prepare the opposite error: the job ID matches, but the example was printed from an earlier wording revision. A release rule that relies only on IDs and names can miss that problem unless there is independent evidence of the actual artifact. Ask the mapper to find one material detail in the visible sample that matches the reviewed version. A complete text check remains appropriate when the job requires it; the demonstration just explains why a label alone isn't proof of content. Approving the exact sample and identifying the artifact are related but not identical acts.

Give the reviewer a way to mark a previous card superseded without returning the old material to a pool of apparently current examples. If the workshop keeps a sample for future comparison, label the revision and disposition with the sample rather than leaving translation to memory. If the paper is later recycled, the receipt can remain as an accurate historical decision without implying that the artifact is still available to inspect. State which details were observed directly and which came from the job record. This lets a later reader notice when the history depends on a document and when it could be independently reproduced using the retained object.

Sources and further reading

AI disclosure: Drafted with AI assistance for Apiary. The sample, workshop and production records are fictional; no production inspection was performed.

Frequently asked
What is Approve the First Piece, Not the Whole Batch about?
Twenty-four signs are due Friday. The design is simple: a small paper placard with a dark border and a space for a holder. On Thursday, somebody looks down at…
What should you know about four different things somebody might call a sample?
A concept sample answers whether people want a thing like this. It can be crude, the wrong material, and wildly unlike the output other than the one feature you want to show. Put a big label on it: “look and layout only” or “prototype; not production.” It's useful at precisely the point where making it…
What should you know about write down what's important before you see anything?
A useful check can fit in a short list. It starts with details that can actually vary. “Does the sign look good?” gives a single observer too much latitude to pass what a second observer would reasonably call wrong. “Finished width 125 millimeters, allowed example band 124 through 126; holder tab must slide into…
What should you know about make the release narrow enough to trust?
The sample card identifies which unit was observed. “One of the blue signs” won't do when they are all blue. Mark the label or sample container prominently, without altering the usable face. Use the job ID, revision and a sample sequence. If you make multiple attempts, keep them distinct: A failed tab fit; B was…
What should you know about worked example: three decisions on the same sign?
Everything here is fictional. One little archive wants twenty-four table labels for a weekend event. The content was approved on PDF P4 at 10:20. The production master M4 has the same wording as P4, plus the verified trim location prescribed for the job. Only one holder fixture fits this run. The lead makes sample A…
References & sources
  1. Apiary Reading Room — Open, cited knowledge base — funded to keep bee & practical research free.
From the Apiary Reading Room. Opinion & editorial — not financial advice. We don't overclaim.
More from the Reading Room